Research

The Learning-enabled Autonomy (LEA) Lab develops learning methods for autonomous and sequential decision-making. Our research brings together reinforcement learning, control theory, optimization, formal methods, and robotics to enable intelligent systems to learn efficiently, reason about complex objectives, and make safe and effective decisions in real-world environments.

Learning Efficiently

We develop reinforcement learning methods that exploit structure in goals, dynamics, representations, and domain knowledge to improve data efficiency and generalization. Our interests include goal-conditioned reinforcement learning, offline reinforcement learning, hierarchical learning, physics-informed learning, and representation learning.

Acting Safely and Effectively

We study how learning-enabled systems can make reliable and effective decisions under uncertainty, particularly in settings where data collection is costly or safety is important. For physical systems, our work connects reinforcement learning with reachability analysis, optimal control, model predictive control, control barrier functions, and optimization.

Reasoning about Complex Objectives

We investigate methods that enable intelligent agents to reason about structured and long-horizon objectives. This includes formal methods, temporal logic, automata, hierarchical decision-making, and learning-based approaches for complex sequential decision problems.

Application Areas

• Robotic manipulation
• Autonomous navigation
• Embodied AI
• Intelligent transportation systems

See Publications for representative work.